Re: [AC21.4] close massive race window on process death, mark II
[moved to -beta, there is no patch here anymore]
On 08 May 2002, Stephen J. Turnbull stipulated:
[Kyle, if you have any time to take a look at this I'd appreciate
your
opinion.]
Seconded. After one embarrassing pratfall I'd appreciate review ;)
nix> Documentation for both `process-live-p' and the new
nix> `process-readable-p' follows, but I guessed you'd want the
nix> revised patch first, because we can live without docs.
Indeed. "But man doth not live by bread alone, but by every word of
Holy Scripture" or something like that.
(Writing 'em now.)
Um ... you said this was "already done in 21.5" ... do you
still think
that's true? Or is it possible that this confusion lives on in the
devel branch?
I thought Ben had torn out the entire process-handling infrastructure in
21.5 and shoved most of it into Lisp.
...
... but, checking, I see that PROCESS_LIVE_P() and friends live on,
although kill-process looks marginally safer.
OK. this needs to be done in 21.5 too, then. I'll do that.
nix> Caveat: I'm not sure whether the call in
create_process()
nix> should be to PROCESS_LIVE_P or PROCESS_READABLE_P. The latter
nix> looks safer, on the assumption that
nix> event_stream_select_process() is a purely I/O and not
nix> signal-based call, so I'm doing that.
This also is the same as the original semantics, which were sorta
kinda reliable.
Yes:) but I frobbed the semantics of some of the other stuff
(set-process-output-coding-system and the like), on the basis that
sending output to a dead process is useless at best and maybe even
harmful (data-loss).
(Er, that is, I frobbed the semantics by not frobbing the code.
Damn, renames are confusing. :) )
--
`Blueshifters were moving away... Redshifters were moving in...'
--- Justina Robson, _Mappa Mundi_, with startling new
evidence for the gib gnab hypothesis